Resource reservation protocol‐traffic engineering (RSVP‐TE) is one of the most widely used technologies to construct multiprotocol label switching tunnels across wide area network. In RSVP‐TE, along the path selected by the constrained shortest path first algorithm, resource is allocated using RSVP at each hop to provide deterministic end‐to‐end service quality, such as rate, delay, packet loss, and jitter. However, we argue that the flat and uniform bandwidth reservation mechanism of RSVP, which reserves the same bandwidth at each hop along the tunnel, ignore the difference of available bandwidth of these links, resulting in that the bandwidth of some congested links are quickly exhausted, whereas other links are not fully utilized. To further improve the load balancing and resource utilization of the network, a nonuniform bandwidth reservation (NUBR) method is proposed in this paper. With the proposed method, each node along the tunnel can flexibly reserve different bandwidth according to its own available resources. Thus, the reserved bandwidth at the congested links can be decreased by aggressively overreserving at idle links. In this paper, we introduce the concept of NUBR and explore the implementation based on the meter table of OpenFlow. The mathematical models and algorithms of NUBR have been studied under 2 application scenarios. At last, the simulation results show that the proposed method can allow more multiprotocol label switching tunnels than the traditional uniform bandwidth reservation method.